[homenet] Comments on src-based routing in general

[Alexandru Petrescu <[e-mail]>]

Hello,

I wanted to make a few informal comments following the J. Chroboczek
presentation of "Source-specific Routing"
draft-boutier-homenet-source-specific-routing-00
today at Berlin IETF Homenet WG mtg.

It is tempting to formalize the prefix and IPv6 addresses into sets and
subsets and relationships between them.  But that relies on an
assumption that 'longest-prefix match' algorithm is precisely known.  It
is not, there is no RFC about it AFAIK.  There are many fuzzy aspects in
the implementation of longest-prefix match (e.g. to disambiguate between
ifaces).

Given that, it is next to impossible to come up with a specification
which is src-based routing (instead of dst-based).

About the problem: I heard said the problem being to ISPs present in the
same homenet, and need to forward the packet to the right ISP based on
the src address.  If that is so, then a first step to realize it would
be to send _some_ traffic to an ISP and the rest to the other.  That
could be still achieved with dst-based routing, but with specific routes
(i.e. for services specific to that ISP use that dst-based specific route).

About the question whether this shows up somewhere else than in
homenets.  One other protocol which uses mandatory src-based routing is
Proxy Mobile IPv6 (the MAG needs it).  One would make sure that way of
doing src-based routing is compatible to homenet.

Two-cents worth,

Alex